Plan: ADR-0009 records Jido v2

docs/adr/0009-jido-for-the-action-effect-layer.md

@@ -1,5 +1,5 @@
 # ADR-0009 — Jido 2.0 for actions, directives and the effect boundary; decided by the Slice 012 design checkpoint
-Status: proposed · Date: 2026-09-05
+Status: accepted · Date: 2026-09-05 · Owner decision on the version line recorded 2026-09-20
 
 ## Context
 ADR/tech-stack v1 said "Jido not chosen". Reconsidered: Jido has a worked vocabulary for actions, directives and
@@ -19,3 +19,18 @@ finalised.
 - Slice 020's tool behaviour becomes a thin wrapper over `Jido.Action` if adopted; the registry rules (compile-time
   catalog, runtime tools cannot enter it) are unchanged.
 - VERSIONS.md gains a `jido ~> 2.0` row pending the checkpoint.
+
+## Decision, appended 2026-09-20
+
+Owner decision, before 2026-09-19, recorded here on 2026-09-20: **the runtime is Jido v2**, pinned `~> 2.3`
+(2.3.3 on hex.pm as of this record; the v3 line is at 3.0.0-beta.1, 2026-09-14, and defers live code migration
+and distributed control-plane claims by its own release notes). The provisional decision above is confirmed on
+the version question and on the shape: actions, directives and the effect boundary use Jido's vocabulary; the
+Session stays an OTP `gen_statem`; the compile-time effect catalog and the rule that runtime tools never enter it
+are unchanged whichever way `Jido.Agent` lands. The Slice 012 design checkpoint still measures (a), (b) and (c)
+as written and reports; what it can no longer return is "reject", because the runtime question is decided. It
+may still return "actions and directives only".
+
+One consequence for the standards register rather than for a slice: another system in the same platform family
+runs on the same Jido line. That is a shared library, not a shared runtime, and it is not on the path a finding
+takes; the register carries the row and the argument, and no slice here does.